reactivated problem escalation code. moved $CRIT_WARN_LEVEL to spong.conf file.

check_tcp used increasing timeouts instead of flat 10 seconds. check_tcp will
now report if service down is due to connection refused or timeout

diff --git a/src/spong-network.pl b/src/spong-network.pl
index d5ef92e92cd9d9463bbbbd694982e89e6fb5cb50..9c00b217f5f1fbcef9654a2d30f7c1963d33405c 100755 (executable)
--- a/src/spong-network.pl
+++ b/src/spong-network.pl
@@ -23,8 +23,8 @@ use lib "@@LIBDIR@@";
 $0 = "spong-network";
 
 use Sys::Hostname;
-use Net::Ping;
 use Socket;
+use IO::Socket;
 use POSIX;
 use Getopt::Long;
 
@@ -32,11 +32,14 @@ use Spong::Daemon;
 use Spong::Status qw(status);
 use Spong::Log;
 
+# Check to see if the Time::HiRes module is available
+eval "use Time::HiRes qw( time );";
+if (! $@ ) { $hires = 1; } else { $hires = 0; };
+
 srand( time() ^ ($$ + ($$ << 15 )) );
 
 $debug = $restart = $kill = 0;
 
-#Getopt::Long::Configure('pass_through');
 if ( ! GetOptions("debug=i" => \$debuglevel, "restart" => \$restart, 
           "kill" => \$kill, "nosleep|refresh" => \$nosleep ) ) {
    &usage();
@@ -157,19 +160,22 @@ sub do_check {
 sub check_simple {
    my( $host, $port, $send, $check, $service ) = @_;
    my( $color, $summary ) = ( "red", "" );
-   my( $attempt, $start, $message, $diff );
+   my( $attempt, $start, $message, $diff, $errcd );
 
-   for $a ( 1..3 ) {
-      $start   = time();
-      $message = &check_tcp( $host, $port, $send );
-      $diff    = time() - $start;
+   for $timeout ( 3, 5, 12 ) {
+      $start = main::time();
+      ($errcd,$message) = &check_tcp( $host, $port, $send, $timeout );
+      $diff    = main::time() - $start;
 
-      $attempt = $a;
+      $attempt++;
       if( $message =~ /$check/ ) { $color = "green"; last; }
       &debug("Attempt $a failed.");
    }
 
-   $summary = "$service down"                            if $color eq "red";
+   $diff = sprintf("%.2f",$diff);
+
+   $summary = "$service is down, timed out" if $color eq "red" and $errcd == 3;
+   $summary = "$service is down, connection refused" if $color eq "red" and $errcd == 2;
    $summary = "$service ok - $diff second response time" if $color eq "green";
    $summary .= ", attempt $attempt" if ($attempt != 1 && $color eq "green");
       
@@ -188,38 +194,47 @@ sub check_simple {
 # ---------------------------------------------------------------------------
 
 sub check_tcp {
-   my( $addr, $port, $data ) = @_;
-   my( $iaddr, $paddr, $proto, $line, $ip );
+   my( $addr, $port, $data, $timeout ) = @_;
+   my( $iaddr, $paddr, $proto, $line, $ip, $sock, $err );
 
    if( $addr =~ /^\s*((\d+\.){3}\d+)\s*$/ ) {
       $ip = $addr;
    } else {
       my( @addrs ) = (gethostbyname($addr))[4];
+      if ( ! @addrs ) { return ( 1, "" ); }
       my( $a, $b, $c, $d ) = unpack( 'C4', $addrs[0] );
       $ip = "$a.$b.$c.$d";
    }
 
-   $iaddr = inet_aton( $ip )                    || return -1;
-   $paddr = sockaddr_in( $port, $iaddr );
-   $proto = getprotobyname( 'tcp' );
- 
    # Set an alarm so that if we can't connect "immediately" it times out.
    # Poor man's exception handling in perl...
 
-   $SIG{'ALRM'} = sub { die };
-   alarm(10);
-
-   eval <<'_EOM_';
-   socket( SOCK, PF_INET, SOCK_STREAM, $proto ) || return -2;
-   connect( SOCK, $paddr )                      || return -3;
-   select((select(SOCK), $| = 1)[0]);
-   print SOCK "$data";
-   recv( SOCK, $line, 256, 0 );  # just grab a chunk from the service.
-   close( SOCK )                                || return -4;
-_EOM_
-   
-   alarm(0);
-   return $line;
+   $err = 0;
+   $line = 0;
+   {
+      local $SIG{'ALRM'} = sub { die "Socket timed out"; };
+      alarm($timeout);
+
+      eval { 
+         $sock = IO::Socket::INET->new( PeerAddr => $ip,
+                                        PeerPort => $port,
+                                        Proto    => 'tcp',
+                                        Reuse    => 1,
+                                   );
+         if ( defined $sock ) {
+            print $sock "$data";
+            $sock->recv( $line, 256, 0 );  # just grab a chunk from the service.
+         } else {
+            $err = 2;  # Could not connect to service port
+         }
+
+         alarm(0);
+      };
+   };
+
+   if ( $@ =~ /timed out/ )  { $err = 3; }
+
+   return ($err, $line);
 }